2.1 Moro background

Moro is classified as a Heiban language within Kordofanian (Schadeberg Reference Schadeberg1981). It has six or seven dialects; the data in this paper are from the Thetogovela dialect (also known as Tobə‹elda), and were collected from two speakers residing in the United States, and one residing in Omdurman, Sudan. The native name for Moro in this dialect is [ðəmwaɾə́ŋá]. Descriptions of other dialects include Black & Black (Reference Black and Black1971) on the ‘Umdorein dialect’, which is probably the dialect referred to as Werria, and forms the basis for standard, written Moro. There is also some discussion of Moro dialects in unpublished material collected by Elizabeth Guest.Footnote ⁶

The obstruent inventory of Moro is given in (3). Obstruents contrast for voicing, but not other laryngeal features. Two fricatives have no contrastive counterparts: voiceless [s] and voiced [ð].

1. (3)

   

The voiceless consonants and [ð] may appear as geminates (sonorants may also be geminated). We will show that when geminated, voiced obstruents (except /ð/) undergo devoicing.

2.2 Dissimilation case 1: the locative prefix

The locative prefix in Moro is [é-] before consonant-initial stems, but [ék-] before vowel-initial stems of the g- noun class and [és-] before vowel-initial stems of the j- noun class. The /e/ surfaces as [i] under vowel harmony. The g- noun class are singular nouns that historically had initial *g and pair with l- class or n- class plurals (Gibbard et al. Reference Gibbard, Rohde, Rose, Matondo, Laughlin and Potsdam2009).Footnote ⁷ The locative marker /ék-/ undergoes dissimilation preceding a voiceless consonant across a vowel (in a rule-based format: /k/→[g] / _VC[–voice]). The /és-/ marker does not undergo dissimilation to [z]; there is no [z] in Moro.

The examples in (4) illustrate locative /ék-/ before g- class vowel-initial nouns that lack voiceless consonants. The addition of the locative prefix also causes high-tone spreading, reduction of some vowels following the prefix and elision of some final vowels; we do not analyse these effects here.

1. (4)

   

If a vowel-initial noun of the g- class has a voiceless consonant following the initial vowel, dissimilation occurs. With nouns of this form, the locative prefix appears as [ég-] or [íg-] rather than [ék-] or [ík-], as shown in (5). Not much data is available for dissimilation before fricatives, as voiceless fricatives are infrequent sounds in Moro. In the one example we found, ‘spirit’ in (5), speakers were unsure whether the prefix should have [k] or [g], which may suggest that fricatives only optionally trigger the dissimilation, or that examples are too rare for speakers to be certain.

1. (5)

   

The dissimilation is subject to a CVC proximity restriction: prefixal /k/ only dissimilates when a voiceless consonant immediately follows the root-initial vowel. Thus the voiceless consonants in (6) do not trigger dissimilation.

1. (6)

   

This kind of proximity restriction is well attested in previous surveys of dissimilation, as well as assimilation (Rose & Walker Reference Rose and Walker2004, Hansson Reference Hansson2010, Bennett Reference Bennett2015).

Two further remarks are in order before moving on to other morphological constructions that exhibit the same phonological pattern. First, the data clearly shows the voicing of /ék-/ → [ég-] to be dissimilatory. It is not the case, for instance, that the consonant is simply voiced intervocalically: words like [logopájá] ‘cup’, [lókógóŋ] ‘scorpion’, [ebamba] ‘drum’, [opéréə] ‘sword’ and [lapaba] ‘circular mat for fanning fire’ clearly show contrastive voicing in the intervocalic position. Second, this dissimilation is clearly for voicelessness, not voicing. Voiced stops may co-occur in adjacent syllables, and do not dissimilate, e.g. [ibəgwɜ́gá] ‘back of knee (instr)’, [ɜ́dígá] ‘skin (instr)’, [égabəɾó] ‘I touched’. Voicing is also not otherwise predictable, e.g. in an alternating voiced–voiceless rhythm. These alternative, non-dissimilatory, characterisations of the data do not fit the facts. We turn now to further data that demonstrates the generality of the phenomenon.

2.3 Dissimilation case 2: verbal suffixes

Dissimilation is not limited to the locative prefix; it is also found with other morphemes. One group is formed by verbal suffixes with the shape /-Vt̪/. These include the benefactive applicative suffix /-ɘt̪/,Footnote ⁸ the inchoative /-et̪/ and the locative applicative /-at̪/; they cause dissimilation of a voiceless consonant in the root, i.e. /T-Vt̪/ → [D-Vt̪].

The benefactive applicative suffix triggers vowel harmony, which raises all vowels in the stem, and causes final dental stops to palatalise; due to dissimilation, these are realised as [ᵈᶾ] rather than [ʧ]. It is clear that this is due to dissimilation, as the passive suffix /-ɘn/ also triggers palatalisation, but here final voiceless dental stops are realised as [ʧ]: e.g. [l-ɜ-wɜʧ-ɘn-ú] ‘they were sewn’. The forms in (7) are conjugated in the 3rd plural perfective with the structure sm.cl1-rtc-root-(appl)-pfv.

1. (7)

   

The examples in (7a) show the applicative suffix in a form with no voiceless consonants, where it is realised as [-ɘt̪]. In the examples in (b) it causes dissimilation in a preceding voiceless consonant. Here, as in the locatives, dissimilation is subject to a CVC proximity restriction, illustrated by the non-dissimilation of root-initial /p/ and /k/ in (c). In applicative forms, we can observe dissimilation applying to consonants other than /k/. Dissimilation in both the applicative and the nominal locative in §2.2 holds in a strictly right-to-left fashion. There are two exceptions to the pattern in regular roots, [l-ɜ-murkw-ɘt̪-ú] ‘they rolled (for)’ and [l-ɜlɘf-ɘt̪-ú] ‘they promised (to)’. The first case may be due to the labial [w] intervening between the voiceless root consonant and the suffix, disrupting locality. The second may be due to fricatives not participating in the dissimilation process as targets. It is an Arabic loanword, so may also be an exception for this reason.

Dissimilation can also be seen with a suffix that we label ‘inchoative’, realised as [-ət̪] or [-et̪], which causes dissimilation in preceding roots. This morpheme appears in imperative forms derived from adjectives, as in (8a). It is [-et̪] if the root is short and contains a short [ə] or [ɘ], and [-ət̪] if the root contains a full vowel or no vowel. We assume that the form with [ə] is a reduced form of the suffix with a full vowel. The form in (b) shows it causing dissimilation in the same fashion as the applicative suffix /-ɘt̪/ above.

1. (8)

   

The locative applicative/malefactive suffix /-at̪/ triggers dissimilation, but unlike the benefactive applicative, it causes no other phonological changes (such as vowel harmony or palatalisation), as shown in (9).

1. (9)

   

The two applicative suffixes /-ɘt̪/ and /-at̪/ cannot combine, so no dissimilation effect can be observed between them. Instead, benefactive + locative applicative meaning is conveyed with a single suffix, /-it̪/, which also triggers dissimilation, e.g. [l-ɜb-it̪-ú] ‘they carried into for’.

2.4 Dissimilation case 3: the reduplicative durative/iterative prefix

The final morphological context where we can observe dissimilation is a durative/iterative prefix. This prefix takes the shape /CaC-/ before roots that are consonant-initial. The consonants of the prefix are copies of the root-initial consonant. The second consonant forms a geminate with the initial root consonant; the vowel is [a] or [ɜ], depending on vowel harmony. Examples are given in (10), with the reduplicant in bold. Labialisation may or may not be copied, depending on the speaker.

1. (10)

   

The /CaC-/ prefix has a [DaT-] pattern with obstruents, as in (11). The result is a voiced–vowel–voiceless sequence: a structure that looks conspicuously like the expected outcome of dissimilation.

1. (11)

   

Two alternations are discernable here: devoicing of geminates, and voicelessness dissimilation. Moro systematically disallows voiced geminate obstruents, with the exception of [ðð], and these are repaired by devoicing, as in (11a): for example, root-initial /b/ in /bə́ɾ/ changes to [p] in [-báp-pəɾ-]. Thus geminates are created by the last consonant of the prefix copying the root-initial consonant, and these are all predictably voiceless, irrespective of the underlying voicing of the root-initial consonant. Further support for the existence of a ban against voiced obstruent geminates comes from the fact that voiceless geminate consonants fail to voice when followed by a voiceless obstruent in a CVC configuration, normally the locus of dissimilation. For example, the inchoative suffix [-et̪] in [mə́ttét̪ áŋəno] ‘be smooth!’ fails to trigger dissimilation: *[mə́ddét̪ áŋəno].

The forms of interest for the voicelessness dissimilation pattern are those in (11b), which show dissimilation of the first consonant in the reduplicative prefix. This first consonant is predictably voiced on the surface, even though the initial root consonant it copies from is voiceless. This fits with the same dissimilation pattern seen elsewhere: in CVC reduplicants, C₂ is always voiceless, which causes C₁ to dissimilate from voiceless to voiced. That is, perfectly faithful copying in the final form in (b) would lead to *[lakákkəva]; but what we find is instead [lagákkəva], reflecting dissimilation of /k/ to [g].

Finally, the voiced fricative /v/ displays variability, as in (12). This is expected, given the previous data on fricatives from locative nominals. The voiceless fricative /s/ shows no variability and is never realised as voiced [z], which, as we have seen, does not exist in Moro.

1. (12)

   

2.5 Dissimilation in roots?

There is evidence that the dissimilation pattern holds within roots as well. From a database of around 1200 lemmas (noun, adjective and verb roots, adverbs and postpositions; compounds were excluded), 111 occurrences of stops/affricates co-occurring across a vowel (CVC configuration) were noted. Observed/expected (O/E) ratios were calculated to test whether there is underrepresentation of particular combinations (Pierrehumbert Reference Pierrehumbert1993). An O/E ratio less than one indicates underrepresentation. ‘Observed’ is the number of stop/affricate pairs for each voicing combination in a CVC configuration. ‘Expected’ is the probability of each combination of [ɑvoice] stops/affricates co-occurring, based on the individual frequency of each. This is calculated by the actual number of [ɑvoice] stops/affricates in C₁ / total number of CVC stop/affricate combinations × the probability of [ɑvoice] stops/affricates in C₂ × the total number of CVC stop/affricate combinations. This distribution is significant (x ² = 26.029, df = 3, p < 0.0001). Table I shows that voiceless–voiceless combinations are very underrepresented and voiced–voiced combinations are also underrepresented, whereas the combinations of voiceless and voiced are overrepresented, particularly voiceless–voiced.

Table I

Static restrictions on voiceless and voiced non-continuant combinations. O/E ratios are given in parentheses.

However, there is considerable skewing, depending on word-class, as can be seen in Table II.

Table II

Number of transvocalic non-continuant obstruent combinations.

In nouns, the distribution is even among all classes, except for the voiceless–voiceless combination. However, in verbs and adjectives, the voiced–voiceless combination dominates. Of the 34 attested combinations, 28 of them involve a second consonant that is either [t̪] or [ʧ], the likely vestige of a lexicalised extension suffix that would have triggered dissimilation, like the applicative markers. This likely accounts for the overrepresentation of this combination in verbs/adjectives. There are too few patterns in the adverb/postposition class to draw many conclusions.

The five examples of voiceless–voiceless combinations are given in (13) (verbs are provided in the proximal imperative, which has a single suffix [-ó]~[-ú]).

1. (13)

   

These words are reasonable exceptions. Geminate stops are always voiceless and never undergo dissimilatory voicing, so dissimilation is not expected in (13a): [óppə́t̪ó], *[óbbə́t̪ó]. The [pw] in (b) and (c) might be considered insufficiently local, due to the [w]. Another possible explanation is that root-initial consonants resist dissimilation. This would also explain the verb [t̪ét̪ó] in (d). The noun [t̪ét̪ə́m] in (e) has an alternative form [d̪ét̪ə́m], showing vacillation between respecting dissimilation and possibly respecting a preference for root-initial voiceless consonants.

As noted above, the behaviour of fricatives in the dissimilation pattern is indeterminate. Co-occurrence of voiceless fricatives with other voiceless obstruents is uncommon in roots, in part due to the low attestation of voiceless fricatives in general (ten lexemes with /f/ and 25 lexemes with /s/ in the non-ideophonic vocabulary). Of these, seven involve combinations of a voiceless fricative with another voiceless obstruent in a CVC configuration. Five of these are cases in which /s/ precedes a voiceless consonant, but /s/ cannot voice to [z]. Two involve /s/ following other voiceless consonants: [alfásəl] ‘room’, borrowed from Arabic, and the verb [tɜ́sí] ‘shake’. It may be the case that fricatives are just not involved in dissimilation in roots, but the evidence is unclear.

2.6 Morpheme-specific dissimilation patterns

We have presented several cases in which dissimilation is triggered by both verbal suffixes and the root. In this section, we outline cases in which dissimilation does not occur, even though the linear configuration voiceless–vowel–voiceless is present. This is due to morphological or prosodic restrictions on its application.

Dissimilation in nouns occurs only across a prefix–root boundary. In (14a), the nouns /et̪a/ ‘lake’ and /opa/ ‘grandmother’ trigger dissimilation on the locative prefix, as expected. However, they fail to undergo dissimilation from the demonstrative suffixes /-íkki/ and /-íkkɜtíkɜ/ in (b), so *[óbɜ́-kki], for example, is not possible.

1. (14)

   

There is also no dissimilation within the demonstrative suffix itself. That is, the form is [-kkɜtíkɜ], not *[-kkɜdíkɜ]. Vowel harmony obeys similar domain boundaries in Moro. Nominal prefixes harmonise, but suffixes, with the exception of inalienable possessives, do not. These demonstratives can stand alone as independent demonstrative pronouns, but vowel-hiatus resolution (/a + i/ → [ɜ]) shows that they are nevertheless phonologically integrated with the preceding noun when suffixed. The domain of dissimilation in nouns is therefore restricted to the morphological unit [prefix–root].

For verbs, we presented all the suffixes with voiceless consonants in §2.4. Here we briefly discuss two cases of single-consonant voiceless prefixes. First, the consecutive complementiser /t̪-/ appears in phrase-initial position, and never undergoes voicing, e.g. [t̪-a-kəl-a] ‘and then you sliced’. Second, there is a phrase-initial strengthening process that causes the noun-class subject agreement prefix /g-/ to be realised as [k-] in phrase-initial position, even in a potentially dissimilating context. In (15a), the noun-class agreement marker is realised as [g], as it is medial in the sentence/phrase. In (b), it is sentence-initial, and is realised as [k], despite the fact that there is a following [t̪].

1. (15)

   

The phrase-initial strengthening process applies regardless of the fact that it creates a voiceless–vowel–voiceless configuration. Phrase-initial strengthening only applies to the initial noun class concord marker /g-/. Other phrase-initial voiced obstruents, as in the imperative verbs /bə́ɾó/ ‘touch!’ and /gə́ɲó/ ‘kill!’, do not undergo devoicing. These two cases show that phrase-initial voiceless realisation of prefixes trumps dissimilation. Other examples of this kind of initial effect will be discussed and analysed in §4.2.

3.2 Acoustic study of Moro stops

4.1 The core dissimilation pattern

The analysis we propose derives dissimilation from the combination of two main constraints. The first is a Corr constraint that motivates correspondence between similar consonants – in this case, voiceless consonants. This defines the featural ‘scope’ of dissimilation: voiceless consonants are the ones that face pressure to correspond or dissimilate. The second major constraint is one that limits correspondence, to make such correspondence impossible in the conditions where dissimilation occurs. We take this to be a CC·Edge constraint, which prompts would-be correspondents to dissimilate when they straddle the edge of a domain.

Locality is built into the theory by making the correspondence constraints sensitive to the particular locality configuration. In the Moro case, this is a CVC sequence or ‘transvocalic’ domain, identified as a domain of correspondence for many other cases of consonant harmony and dissimilation (Rose Reference Rose2000, Hansson Reference Hansson2010, Bennett Reference Bennett2015). This confines the interaction to adjacent syllables. The participating consonants are highly similar: voiceless stops and affricates, identified by the features [—voice, —continuant] (fricatives are set aside, as their participation is doubtful). If [–voice] were not employed as a defining feature of the natural class, [—continuant] alone would be insufficient: this would predict correspondence among all stops and affricates, and dissimilation for continuancy. Absence of a feature cannot be referred to formally to define a class without invoking a [Ø] feature.

1. (16)

   

The Corr constraint in (16) interacts with another constraint that penalises correspondence across domain edges by requiring correspondence to be within a particular domain: CC·Edge(domain). Domains may be phonological (syllable, foot, prosodic word) or morphological (root, stem, word). This constraint also establishes domains for consonant harmony (see Bennett Reference Bennett2015 for more details). Since the three clear and productive subcases of dissimilation in Moro (discussed in §2 above) each hold across root–affix boundaries, we presume the relevant CC·Edge constraint to be one that picks out the root as its domain, as in (17).

1. (17)

   

To illustrate with the example /ék-etám/ → [égətám] ‘in the neck’, the two surface correspondence constraints are ranked above a faithfulness constraint that prevents changes to the feature [voice]: IO·Ident[voice]. This is shown in (18).Footnote ¹¹

1. (18)

   

Correspondence relationships between surface consonants () are marked in set notation, with each group of correspondents co-indexed to the consonants it includes. If there is total non-correspondence among segments, as in (18a) and (c), then each consonant forms its own set. Corr-CVC·[—voice, —cont] requires voiceless obstruents to correspond. This rules out faithful non-correspondence between two voiceless stops, as in (18c). CC·Edge(root) wants the voiceless consonants here to not correspond; since they are separated by a prefix–root boundary, correspondence between them crosses the edge of the root. This rules out faithful correspondence between the voiceless stops, as in (b). Thus pairs of […voice] consonants are required to correspond by one constraint, but are prohibited from doing so by the other constraint: they cannot both correspond and also satisfy the requirement imposed on correspondence by CC·Edge. The solution is to make them not be a pair of [–voice] Cs: dissimilating one to [+voice] as in (18a) allows both constraints to be satisfied at the same time.

This same basic constraint ranking also applies to cases in which a suffix triggers a change to a root consonant, as in /l-ap-ɘt̪-ó/ → [lɜbɘt̪ú] ‘they carried for someone’. This is because a suffixal voiceless consonant and a root /p/ straddle the root edge, just like the /k-Vt/ sequence shown in (18). Any pair of voiceless stops or affricates that crosses the root edge is compelled to dissimilate under this ranking.

Moro dissimilation is always regressive, so the output in (18) is [égətám], not *[ekədám]. We do not propose an explanation here for this rigid directionality, but simply note that left-to-right dissimilation is systematically not found. Recent work on surface correspondence has viewed directionality as an unsolved problem (see e.g. Rose & Walker Reference Rose and Walker2004, Hansson Reference Hansson2010, Bennett Reference Bennett2015). In the Moro case, a faithfulness constraint that protects the rightmost of any two voiceless consonants would be sufficient to force this right-to-left directionality on the process. Jurgec (Reference Jurgec2011) posits positional faithfulness constraints that pick out the rightmost or leftmost occurrence of a feature in a domain; a constraint of this form, such as IO·Ident[voice]/CVC-Rightmost, would suffice for our analysis.

Now let us turn to reduplicated prefixes. Roots with initial voiceless consonants show dissimilation in the reduplicative prefix – exactly what would be expected from copying the root consonant in the usual way, then applying the dissimilatory alternation in the same fashion as above. In this case, we see dissimilation trump base–reduplicant identity, rather than input–output identity; like IO·Ident, BR·Ident is also ranked lower than the Corr and CC·Edge constraints, as shown in (19). (We take the geminates created through this reduplication, such as [kk] in (19), to be single long consonants which straddle the root–prefix boundary, and indicate their uniformity with a tiebar.)

1. (19)

   

The reduplicative prefix shows a further layer of complexity, because Moro has a ban on voiced geminate obstruents (as noted above in §2.5). If the initial root consonant is voiced, we would expect a voiced geminate to be created through reduplication: reduplicative prefixation of /bəɾ/ ‘touch’ should yield [-báb-bəɾ-]. However, the ban on voiced geminates, formalised here as a constraint *VoiObs-Gem, rules this out in favour of a geminate [pp] instead of [bb]. *VoiObs-Gem is thus ranked higher than BR·Ident[voice], as in (20). Given this ban on voiced geminates, the initial consonant of the reduplicative prefix exhibits a further dissimilatory effect. With /bəɾ/, geminate devoicing and faithful reduplication should give [-páp-pəɾ-], as in (c, d); however, the form we actually observe is [-báp-pəɾ-] in (a), with apparent dissimilation in the reduplicant. This result falls out from the ranking established in (19), together with undominated *VoiObs-Gem.

1. (20)

   

4.2 Resistance to dissimilation

In §2 we briefly discussed three situations in which dissimilation does not apply to a TVT sequence: (i) if the first T is a geminate obstruent, (ii) if the second T is in a nominal suffix and (iii) if the first T is in phrase-initial position. In this section we present two more cases of dissimilation failing to apply in the TVT configuration, ones for which the explanation constitutes an emergent effect. We show that both cases can be attributed to higher-ranked constraints that interact with those given above.

The first case involves the iterative/durative reduplicative prefix when it appears in stems in the imperative. Unlike in other verb forms with this prefix, there is no dissimilation, as shown in (21). In fact, there is an additional requirement that both obstruents be voiceless, no matter the original voicing of the initial root consonant.

1. (21)

   

We propose that there is a preference for initial obstruents to be voiceless rather than voiced, which is manifested in these forms as a resistance to dissimilation. Two known aerodynamic factors conspire towards initial voiceless obstruents: (i) utterance-initial position favours longer closures (Westbury & Keating Reference Westbury and Keating1986), and (ii) the longer the stop closure, the harder it is for voicing to be maintained (Ohala & Riordan Reference Ohala, Riordan, Wolf and Klatt1979). In Moro, voiced stops and affricates in word-initial position are permitted, as we have already seen in the acoustic experiment. However, there is some evidence that there is a gradient tendency to avoid them. Voiced consonants in initial position are less common than voiceless ones both for lexemes overall (70 vs. 104) and within all lexical categories. Initial /g/ has been lost historically in most words. Word-initial /b d̪ d ᵈᶾ/ are rare in nouns (ten attestations), largely because most nouns have a noun class prefix, and none of the class prefixes include these consonants (Gibbard et al. Reference Gibbard, Rohde, Rose, Matondo, Laughlin and Potsdam2009). They may appear word-initially in verbs, but only in imperative forms; in all other contexts they are preceded by prefixes. Moreover, in related Kordofanian languages, a preference for voiceless initial consonants is also reported. In Koalib (Quint Reference Quint2009), only voiceless obstruents are attested word-initially; voiced obstruents appear postvocalically. In Dagik (Vanderelst Reference Vanderelst2016), the distribution is similar: voicing is allophonic, with voiceless stops only appearing word-initially and word-finally (the latter position limited to [k]).

We therefore suggest that a preference for voiceless initial consonants can emerge in reduplication. Unlike word-medial consonants, initial Cs resist dissimilation. This effect is independent of the morphemes involved, and is entirely about the initial position, as is clear from a comparison between [páppə́ɾó] ‘touch!’ and [labáppəɾa] ‘they are about to touch’, both from the root /bəɾ/. The core explanation for these differences is that reduplicated [+voice] stops and affricates are different from underlying ones, as they are subject to BR faithfulness, but not IO faithfulness. Positional IO faithfulness protects underlying initial voiced stops, but allows reduplicated ones to be neutralised.

First, we assume that there is a penalty against initial voiced obstruents: *#[+voice].Footnote ¹² However, a high-ranked positional IO faithfulness constraint protects underlying voiced obstruents in word-initial position from altering their voicing specification. The result is that underlying voiced stops are faithful in initial position, as in (22).

1. (22)

   

Reduplicated consonants, however, are not subject to the positional faithfulness constraint, as it targets the input–output relationship. Since the markedness constraint penalising initial voiced obstruents outranks the surface correspondence constraints (which drive the dissimilation), the result is an initial voiceless consonant, as in (23). This happens to obey base–reduplicant identity, though BR·Ident is not what drives the interaction.Footnote ¹³

1. (23)

   

Second, there is a case of resistance to dissimilation with roots consisting of only a single voiceless consonant. While some of these are geminate consonants, and would be covered by the ‘no voiced obstruent geminates’ generalisation, others are not. Two adjectives of this type, /ʧ/ ‘bad’ and /t/ ‘small’, remain voiceless when a following inchoative suffix is added in the imperative, and when the benefactive applicative is added (with a comparative sense), as in (24a). This shows that it is not their word-initial status in the imperative that favours maintaining their voiceless status. In addition, three verbs, /pw/ ‘beat’, /t̪/ ‘drink’ and /t̪w/ ‘get lost’, also show resistance to voicing with the benefactive applicative, as shown in (24b).

1. (24)

   

This does not apply to the VC root /ap/ ‘carry’. The applicative perfective is [lɜbɘt̪ú] ‘they carried for’ with dissimilation, which can be compared with [lɜpɘt̪ú] in (24), with no dissimilation.Footnote ¹⁴ As such, resisting dissimilation appears to be a property of single-consonant roots, or perhaps more generally of root-initial consonants (all roots are consonant-final, so there are no CV roots).Footnote ¹⁵ We do not develop a full analysis of this here, but note that a markedness or positional faithfulness constraint is required that outranks the standard dissimilation-inducing constraints. We propose that this is a constraint that penalises root-initial voiced obstruents. As such obstruents are attested, this dispreference would only emerge when a root is short enough to fit the CVC configuration required for dissimilation triggered by a suffix – when the root is C only. This analysis has the advantage of connecting to the other cases of dispreference for voiced obstruents in initial positions. In phrase-initial position, /g/ devoices to [k], and /t̪-/ resists dissimilation (§2.6). In word-initial position, voiced obstruents are allowed, but are realised as voiceless when copied in reduplication, which is not subject to general faithfulness.

In conclusion, these two cases highlight the special status of initial word position and initial root position in disallowing voiced obstruents as the result of dissimilation. The cumulative ranking for the full pattern is given in (25).

1. (25)

   

4.3 Dissimilation in the lexicon revisited

In §2.5 we noted the tendency for voiceless–voiceless sequences to be underrepresented in lexical roots. What this observation means for the grammar is not obvious. One interpretation of the facts is that voiceless–voiceless sequences are rare because dissimilation also applies within roots: TVT roots are uncommon because they productively dissimilate to DVT. On this interpretation, dissimilation crucially holds not only across root boundaries, but also within roots.

The analysis developed so far does not predict dissimilation within roots: the combination of CC·Edge(root) and Corr-CVC·[—voice, —cont] can derive dissimilation only across the edge of the root domain. However, it is possible to construct a very similar analysis, also using the SCTD, which does make that prediction. This is done by calling on a different CC·Edge constraint: CC·Edge(σ). In precisely the same way that CC·Edge(root) prohibits correspondence – and accordingly supports dissimilation – when two voiceless consonants straddle the edge of a syllable, CC·Edge(σ) supports dissimilation across the edge of any syllable. As such, if we slot CC·Edge(σ) in place of CC·Edge(root) into the same ranking structure given in (25), the result in (26) is a grammar that enforces voicelessness dissimilation across any syllable edge, modulo the same restrictions on voicing geminates and initial consonants.Footnote ¹⁶

1. (26)

   

By requiring voiceless obstruents in CVC sequences to correspond, and prohibiting correspondence across syllable edges, this ranking causes dissimilation for transvocalic heterosyllabic voiceless consonants, irrespective of morphological structure. The key interaction is illustrated in (27), using a hypothetical input /apoto/ (cf. [ábótó] ‘climb’) to show root-internal dissimilation.Footnote ¹⁷

1. (27)

   

The outcome here is intensionally (and intentionally) different than in the previous version of the analysis, but its extensional effects are largely the same. The morphology of Moro is such that any CVC pair of obstruents that straddles the edge of the root always straddles a syllable boundary too (e.g. [é.g-ə.tám]). So the interaction illustrated here also derives dissimilation in all the morphological contexts considered above.

The point here is that if the root-internal co-occurrence facts are taken to reflect a real dissimilation process happening in the phonology of Moro, then we can explain that process using the same theory and same basic approach as in the analysis laid out above. The situation is that the data does not fully reveal what is happening in the phonological grammar of Moro. Underrepresentation of voiceless–voiceless pairs in roots could be the result of dissimilation in the root (albeit with some lexical exceptions), or it could be that such roots are rare for independent reasons: perhaps voiced–voiceless sequences accumulated over time in the lexicon, while voiceless–voiceless ones did not (à la Coetzee Reference Coetzee2014, for instance). The general approach of our analysis is consistent with either scenario, and both of the resulting patterns can be explained by the SCTD.